Northern high latitudes contain large amounts of soil organic carbon (SOC),
of which Alaskan terrestrial ecosystems account for a substantial proportion.
In this study, the SOC accumulation in Alaskan terrestrial ecosystems over
the last 15 000 years was simulated using a process-based biogeochemistry
model for both peatland and non-peatland ecosystems. Comparable with the
previous estimates of 25–70 Pg C in peatland and 13–22 Pg C in
non-peatland soils within 1 m depth in Alaska using peat-core data, our
model estimated a total SOC of 36–63 Pg C at present, including
27–48 Pg C in peatland soils and 9–15 Pg C in non-peatland soils.
Current vegetation stored 2.5–3.7 Pg C in Alaska, with 0.3–0.6 Pg C in
peatlands and 2.2–3.1 Pg C in non-peatlands. The simulated average rate of
peat C accumulation was 2.3 Tg C yr
The P-TEM (Peatland-Terrestrial Ecosystem Model) framework includes a soil thermal module (STM), a hydrologic module (HM), a carbon/nitrogen dynamic model (CNDM), and a methane dynamics module (MDM) (Wang et al., 2016).
Global surface air temperature has been increasing since the middle of the
19th century (Jones and Mogberg, 2003; Manabe and Wetherald, 1980, 1986).
Since 1970, the warming trend has accelerated at a rate of
0.35
The warming climate could increase C input to soils as litters by stimulating plant net primary productivity (NPP) (Loisel et al., 2012). However, it can also decrease the SOC by increasing soil respiration (Yu et al., 2009). Warming can also draw down the water table in peatlands by increasing evapotranspiration, resulting in higher decomposition as the aerobic respiration has a higher rate than anaerobic respiration in general (Hobbie et al., 2000). SOC accumulates where the rate of soil C input is higher than decomposition. The variation of climate may switch the role of soils between a C sink and a C source (Davidson and Janssens, 2006; Davidson et al., 2000; Jobbágy and Jackson, 2000). Unfortunately, due to the data gaps of field measurement and uncertainties in estimating regional C stock (Yu, 2012), with limited understanding of both peatlands and non-peatlands and their responses to climate change, there is no consensus on the sink and source activities of these ecosystems (Frolking et al., 2011; Belyea, 2009; McGuire et al., 2009).
Description of sites and variables used for parameterizing the core carbon and nitrogen module (CNDM).
Both observation and model simulation studies have been applied to understand the long-term peat C accumulation in northern high latitudes. Most field estimations are based on series of peat-core samples (Turunen et al., 2002; Roulet et al., 2007; Yu et al., 2009; Tarnocai et al., 2009). However, those core analyses may not be adequate for estimating the regional C accumulation due to their limited spatial coverage. To date, a number of model simulations have also been carried out. For instance, Frolking et al. (2010) developed a peatland model considering the effects of plant community, hydrological dynamics and peat properties on SOC accumulation. The simulated results were compared with peat-core data. They further analyzed the contributions of different plant functional types (PFTs) to the peat C accumulation. However, this 1-D model has not been evaluated with respect to soil moisture, water-table depth, methane fluxes, and carbon and nitrogen fluxes and has not been used in large spatial-scale simulations by considering other environmental factors (e.g., temperature, vapor pressure, and radiation). In contrast, Spahni et al. (2013) used a dynamic global vegetation and land surface process model (LPX), based on LPJ (Sitch et al., 2003), imbedded with a peatland module, which considered the nitrogen feedback on plant productivity (Xu-Ri and Prentice, 2008) and plant biogeography, to simulate the SOC accumulation rates of northern peatlands. However, climatic effects on SOC were not fully explained, presumably due to its inadequate representation of ecosystem processes (Stocker et al., 2011, 2014; Kleinen et al., 2012). The Terrestrial Ecosystem Model (TEM) has been applied to study C and nitrogen dynamics in the Arctic (Zhuang et al., 2001, 2002, 2003, 2015; He et al., 2014). However, the model has not been calibrated and evaluated with peat-core C data, and has not been applied to investigate the regional peatland C dynamics. Building upon these efforts, recently we fully evaluated the peatland version of TEM (P-TEM) including modules of hydrology (HM), soil thermal (STM), C and nitrogen dynamics (CNDM) for both upland and peatland ecosystems (Wang et al., 2016).
Here we used the peatland-core data for various peatland ecosystems to parameterize and test P-TEM (Fig. 1). The model was then used to quantify soil C accumulation of both peatland and non-peatland ecosystems across the Alaskan landscape since the last deglaciation. This study is among the first to examine the peatland and non-peatland C dynamics and their distributions and peat depths using core data at regional scales.
To conduct regional simulations of carbon accumulation for both uplands and peatlands, we first parameterized the P-TEM for representative ecosystems in Alaska. Second, we organized the regional vegetation and peatland distribution data, spatial basal age data for all peatland grid cells based on site-level soil core data, and climate data for each period during the Holocene. Finally, we conducted the regional simulations and sensitivity analysis.
In P-TEM (Wang et al., 2016), peatland soil organic C (SOC) accumulation is
determined by the difference between NPP and aerobic and anaerobic
decomposition. Peatlands accumulate C where NPP is greater than
decomposition, resulting in positive net ecosystem production (NEP):
We modeled peatland soils as a two-layer system for a hydrological module (HM) while keeping the three-layer system for upland soils (Zhuang et al., 2002). The soil layers above the lowest water-table position are divided into (1) a moss (or litter) organic layer (0–10 cm) and (2) a humic organic layer (10–30 cm) (Wang et al., 2016). Based on the total amount of water content within those two unsaturated layers, the actual water-table depth (WTD) is estimated. The water content at each 1 cm above the water table can then be determined after solving the water balance equations (Zhuang et al., 2004).
Carbon pools and fluxes used for calibration of CMDM.
In the STM module, the soil vertical profile is divided into four layers: (1) snowpack in winter, (2) a moss (or litter) organic layer, and (3) upper and (4) lower humic organic soil (Wang et al., 2016). Each of these soil layers is characterized by a distinct soil thermal conductivity and heat capacity. We used the observed water content to drive the STM (Zhuang et al., 2001).
The methane dynamics module (MDM) (Zhuang et al., 2004) considers the processes of methanogenesis, methanotrophy, and the transportation pathways, including (1) diffusion through the soil profile, (2) plant-aided transportation and (3) ebullition. The soil temperatures calculated from STM, after interpolation into 1 cm sub-layers, are input to the MDM. The water-table depth and soil water content in the unsaturated zone for methane production and emission are obtained from HM, and NPP is calculated from the CNDM. Soil-water pH is prescribed from observed data and the root distribution determines the redox potential (Zhuang et al., 2004).
We have parameterized the key parameters of the individual modules, including
HM, STM, and MDM in Wang et al. (2016). The parameters in CNDM for upland
soils and vegetation have been optimized in the previous studies (Zhuang et
al 2002, 2003; Tang and Zhuang, 2008). Here we parameterized P-TEM for
peatland ecosystems using data from a moderately rich
Alaskan vegetation distribution maps reconstructed from fossil
pollen data during
Assignment of biomized fossil pollen data to the vegetation types in TEM (He et al., 2014).
Relations between peatland basal age and vegetation distribution.
The Alaskan C stock was simulated through the Holocene driven with vegetation
data reconstructed for four time periods, including a time period
encompassing a millennial-scale warming event during the last deglaciation
known as the Bølling–Allerød at 15–11 ka (1 ka
The upland and peatland ecosystem distribution for each grid cell was
determined using the wetland inundation data extracted from the NASA/GISS
global natural wetland dataset (Matthews and Fung, 1987). The resolution was
resampled to
Our regional simulations considered the effects of basal ages on carbon accumulation. To obtain the spatially explicit basal age data for all peatland grid cells, we first categorized the observed basal ages of peat samples from Gorham et al. (2012) into different time periods, including 15–11, 11–10, 10–9, and 9 ka–19th (Fig. 2). For each time period, the areas dominated by different vegetation types were assigned with varying peatland basal ages. To do that, we examined the association of peat basal ages and vegetation types from peat-core data. For instance, we found that peatland initiations during 15–11 ka occurred in the regions that were dominated by alpine tundra at the southern, northwestern, and southeastern coasts. We thus assign the different peatland basal ages for the grid cells according to their vegetation types for each time slice (Table 4).
Climate data were bias-corrected from ECBilt-CLIO model output (Timm and
Timmermann, 2007) to minimize the difference from CRU data (He et al., 2014).
Climate fields include monthly precipitation, monthly air temperature,
monthly net incoming solar radiation, and monthly vapor pressure at a
resolution of
Sites used for comparison of carbon accumulation rates between simulation and observation (Jones and Yu, 2010).
Simulated paleo-climate and other input data from 15 ka to
2000 AD:
Simulations for pixels located on the Kenai Peninsula from 15 to 5 ka were first conducted with the parameterized model. The peat-core data from four peatlands on the Kenai Peninsula, Alaska (Jones and Yu, 2010; Yu et al., 2010) (Table 5; also see Table 3 in Wang et al., 2016) were used to compare with the simulations. The observed data include the peat depth, bulk density of both organic and inorganic matters at 1 cm interval, and age determinations. The simulated C accumulation rates represent the actual (“true”) rates at different times in the past. However, the calculated accumulation rates from peat cores are considered “apparent” accumulation rates, as peat would continue to decompose since the time of formation until the present when the measurement was made (Yu, 2012). To facilitate comparison between simulated and observed accumulation rates, we converted the simulated “true” accumulation rates to “apparent” rates, following the approach by Spahni et al. (2013). That is, we summed the annual net C accumulation over each 500-year interval and deducted the total amount of C decomposition from that time period, then dividing by 500 years.
Second, we conducted a transient regional simulation driven with monthly
climatic data (Fig. 3) from 15 ka to 2000 AD. The simulation was conducted
assuming all grid cells were taken up by upland ecosystems to get the upland
soil C spatial distributions during different time periods. We then conducted
the second simulation assuming all grid cells were dominated by peatland
ecosystems following Table 3 to obtain the distributions of peat SOC
accumulation. Finally, we used the inundation fraction map to extract both
uplands and peatlands and estimated the corresponding SOC stocks within each
grid, which were then summed up to represent the Alaskan SOC stock. We also
used the observed mean C content of 46.8 % in peat mass and bulk density
of 166
Third, we conducted a series of extra simulations to further examine how
uncertain climates and vegetation distribution affect our results. We used
the original forcing data as the standard scenario and the warmer (monthly
temperature
Simulated and observed carbon accumulation rates from 15 to 5 ka in
20-year bins
Our paleosimulations showed a large peak of peat C accumulation rates at
11–9 ka during the HTM (Fig. 4). The simulated “true” and “apparent”
rates captured this primary feature in peat-core data at almost all sites
(Jones and Yu, 2010; see Wang et al., 2016, Table 3 for site details). We
simulated an average of the peat SOC “apparent” accumulation rate of
11.4 g C m
Simulated
Total C (Pg C) stored in Alaskan vegetation for different time periods.
Model simulations showed an overall low vegetation C before the HTM
(15–11 ka) (Fig. 5a) parallel to the relatively low annual and long-term
NPP (Fig. 5b and c). The lowest amount of C
(
In general, vegetation held about 2 Pg C before the HTM (Fig. 6). Upland tundra ecosystems accounted for the highest amount of C. During the HTM, boreal evergreen needleleaf forest reached its peak and had an overwhelming proportion over total C. Similarly, a peak of total vegetation C appeared at the same time, averaging around 4.3 Pg C. A large decrease occurred at the mid-Holocene and a slight decline continued till the late Holocene. We estimated a total of 2.9 Pg C stored in modern Alaskan vegetation, with 0.4 Pg in peatlands and 2.5 Pg in non-peatlands. The uncertainties during the model calibration (Table 2) resulted in 0.3–0.6 and 2.2–3.1 Pg C in peatlands (see Wang et al., 2016, for model parameters) and non-peatland vegetation (see Tang and Zhuang, 2008, for uncertainty analyses for upland vegetation), respectively. Our estimation of 2.5–3.7 Pg C stored in the Alaskan vegetation was lower than the previous estimate of 5 Pg (Balshi et al., 2007; McGuire et al., 2009), presumably due to the prior ranges of model parameters used from Tang and Zhuang (2008). Our overestimation of peatland area may also lead to a reduction of Alaskan non-peatland area.
Average non-peatland (mineral) SOC density (kg C m
Peatland area expansion and peat soil C accumulation per 1000 years
(kg C m
Carbon storage in Alaskan non-peatland soils varied spatially (Fig. 7). Moist
tundra had the highest SOC density (12–25 kg C m
Peatland expansion area (
An average peat SOC “apparent” accumulation rate of
13 g C m
Bars of peatland mean C accumulation rates from 15 ka to 2000 AD
for
Total C stock accumulated from 15 ka to 2000 AD for all peatlands,
The SOC stock of northern peatlands has been estimated in many studies,
ranging from 210 to 621 Pg (Oechel, 1989; Gorham, 1991; Armentano and
Menges, 1986; Turunen et al., 2002; Yu et al., 2010; see Yu, 2012, for a
review). Assuming Alaskan peatlands were representative of northern peatlands
and using the area of Alaskan peatlands (
The northern circumpolar soils were estimated to cover approximately
Spatial distribution of
The simulated modern SOC distribution (Fig. 12c) was largely consistent with
the study of Hugelius et al. (2014) (see Fig. 3 in the paper). The model
captured the SOC density on the northern and southwestern coasts of Alaska,
with most grids
The simulated climate by the ECBilt-CLIO model showed that among the six time periods, the coolest temperature appeared at 15–11 ka, followed by the mid and late Holocene (5 ka–1900 AD). Those two periods were also generally dry (Fig. 3f). The former represented colder and drier climates before the onset of the Holocene and the HTM (Barber and Finney, 2000; Edwards et al., 2001). The latter represented post-HTM neoglacial cooling, which has caused permafrost aggradation across northern high latitudes (Oksanen et al., 2001; Zoltai, 1995).
Field-based estimates and model simulations for peat depths in
Alaska: the observed and simulated data are extracted from the same grids on
the map. The linear regression line (cyan) is compared with the 1 : 1 line. The
linear regression is significant (
Temperature and precipitation effects on
Despite the relatively large inter-annual NPP variation that resulted from
the annual fluctuations of the climate forcing (Fig. 5b), the long-term NPP,
vegetation C density and storage were highest during the HTM (Fig. 5a and c).
Annual C accumulation rates also reached their peak (Figs. 5–11). The
long-term variation of NPP has a similar pattern of the climate (see Fig. 3
for climate variables), where higher NPP, along with higher vegetation C,
coincided with warmer temperatures and enhanced precipitation during the HTM
compared to other time periods. ECBilt-CLIO simulated a warmest summer and a
prolonged growing season, leading to a stronger seasonality of temperature
during the HTM (Kaufman et al., 2004, 2016), in line with the orbitally
induced maximum summer insolation (Berger and Loutre, 1991; Renssen et al.,
2009). The coincidence between the highest vegetation C uptake and SOC
accumulation rates and the warmest summer and the wetter-than-before
conditions indicated a strong link between those climate variables and C
dynamics in Alaska. Enhanced climate seasonality characterized by warmer
summer, enhanced summer precipitation and possibly earlier snowmelt during
the HTM accelerated the photosynthesis and subsequently increased NPP (Tucker
et al., 2001; Kimball et al., 2004; Linderholm, 2006). As shown in our
sensitivity test, annual NPP was increased by 40 and
20 g C m
Climate variables significantly affect C dynamics within each time slice. However, different vegetation distributions during various periods led to clear step changes, suggesting vegetation composition was likely to be another primary factor (Figs. 6, 7, 8, and 11). As key parameters controlling C dynamics in the model (e.g., maximum rate of photosynthesis, litter fall C, maximum rate of monthly NPP) are ecosystem type specific, vegetation distribution changes may drastically affect regional plant productivity and C storage. Our sensitivity test indicated that by replacing all vegetation types with forests, there was a total increase of 36.9 Pg in upland plus peatland soils. There was also an increase of 48.8 Pg C under warmer and wetter conditions, suggesting that both climate and vegetation distribution may have played important roles in carbon accumulation.
The vegetation changes reconstructed from fossil pollen data during different
time periods followed the general climate history during the last
15 000 years. For instance, the migration of dark boreal forests over
snow-covered tundra during the HTM was probably induced by the warmer and
wetter climate that resulted from the insolation changes (He et al., 2014).
The cooler and drier climate after the mid-Holocene limited the growth of
boreal broadleaf conifers (Prentice et al., 1992), and therefore resulted in
the replacement of broadleaf forest with needleleaf forest and tundra
ecosystems. Since the parameters of our model for individual vegetation type
were static, parameterizing the model using modern site-level observations
might have introduced uncertainty to parameters, which may result in regional
simulation uncertainties. Assuming each parameter as constant (e.g., the
lowest water-table boundary; see Wang et al., 2016, for details) over time
may also weaken the model's response to different climate scenarios.
Furthermore, applying static vegetation maps at millennial scales and using
modern elevation and pH data may simplify the complicated changes in
landscape and terrestrial ecosystems, as vegetation can shift within hundreds
of years (Ager and Brubake, 1985; see He et al., 2014, discussion section).
Relatively coarse spatial resolution (
We used a biogeochemistry model for both peatland and non-peatland ecosystems
to quantify the C stock and its changes over time in Alaskan terrestrial
ecosystems during the last 15 000 years. The simulated peat SOC accumulation
rates were compared with peat-core data from four peatlands on the Kenai
Peninsula in southern Alaska. The model estimated the peat SOC accumulation
rate trajectory throughout the Holocene well. Our regional simulation showed
that 36–63 Pg C had been accumulated in Alaskan land ecosystems starting
15 000 years ago, including 27–48 Pg C in peatlands and 9–15 Pg C in
non-peatlands (within 1 m depth). We also estimated that 2.5–3.7 Pg C was
stored in contemporary Alaskan vegetation, with 0.3–0.6 Pg C in peatlands
and 2.2–3.1 Pg C in non-peatlands. The estimated average rate of peat C
accumulation was 2.3 Tg C yr
Data presented in this paper are publicly accessible: ECBilt-CLIO
Paleosimulation (
We acknowledge the funding support from NSF project IIS-1027955 and DOE project DE-SC0008092. We also acknowledge the SPRUCE project for allowing us to use its data. Edited by: A. V. Eliseev Reviewed by: two anonymous referees